Looking for Alzheimer's Disease morphometric signatures using machine learning techniques

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning

Background:

• Automated prediction of Mild Cognitive Impairment (MCI) from MRI data is crucial for early diagnosis.
• Neuromorphometrics features (nMF) derived from MRI show potential in classifying cognitive states.
• Distinguishing between Healthy Controls (HC), MCI, converters MCI (cMCI), and Alzheimer's Disease (AD) patients is a key challenge.

Purpose of the Study:

• To evaluate the performance of MRI-based nMF in classifying HC, MCI, cMCI, and AD patients.
• To develop and assess a Multi Classifier System (MCS) for improved diagnostic accuracy.
• To investigate the impact of cognitive scoring (MMSEs) on feature selection and classification performance.

Main Methods:

• Participants were grouped based on Mini Mental State Examination scores (MMSEs).
• Key nMF were identified for each cognitive group.
• A Multi Classifier System (MCS) was developed using these selected nMF and compared against single classifier approaches.
• Three state-of-the-art classification algorithms were employed for comparison.

Main Results:

• The MCS demonstrated superior performance in Accuracy and Area Under the Receiver Operating Curve (AUC) compared to single classifiers.
• Multiclass AUCs for the MCS were 0.83 (HC), 0.76 (cMCI), 0.65 (MCI), and 0.95 (AD).
• The MCS achieved 81.0% accuracy (AUC=0.88) for Neurodegenerative Disease (ND) detection, outperforming single classifiers (71.3% and 63.1%).

Conclusions:

• The proposed MCS significantly outperforms single classifier systems in classifying cognitive impairment stages.
• Integrating cognitive scoring (MMSEs) into MCS design enhances the selection of relevant MRI-based features.
• This study highlights the potential of MCS combined with MRI and cognitive data for accurate neurodegenerative disease prediction.